Just shipped OSS Sahayak for Hackhazards 2026 — solo build 🚀
The problem: maintainers can't tell from a PR alone whether a contributor actually explored the codebase, or just produced something that looks competent. OSS Sahayak fixes that by reconstructing engineering context directly from a repo's git history.
For contributors: a deterministic Learning Path through the relevant code before you start, plus evidence-backed Q&A (with Hinglish/voice support).
For maintainers: a Contribution Readiness Report — Exploration Coverage, Evidence Coverage, Dependency Coverage — computed transparently from real graph data, never an AI's subjective guess.
Tech stack:
- Neo4j AuraDB as the persistent evidence graph (8 node types, 10 relationship types)
- Gemini as the sole reasoning layer
- Sarvam AI (Saaras STT, Sarvam-30B/105B, Bulbul TTS) for the accessibility/Hinglish layer
- Render Workflows for a durable, checkpointed, incrementally-resumable ingestion pipeline
Every metric shown in the product is computed from real ingested data — no mock numbers, no placeholders.
Top comments (0)